Truncated Stathmin-2 Is a Marker of TDP-43 Pathology in Frontotemporal Dementia

Truncated stathmin-2 is a marker of TDP-43 pathology in frontotemporal dementia

Mercedes Prudencio, … , Pietro Fratta, Leonard Petrucelli

J Clin Invest. 2020;130(11):6080-6092. https://doi.org/10.1172/JCI139741.

Research Article Neuroscience

No treatment for frontotemporal dementia (FTD), the second most common type of early-onset dementia, is available, but therapeutics are being investigated to target the 2 main proteins associated with FTD pathological subtypes: TDP-43 (FTLD-TDP) and tau (FTLD-tau). Testing potential therapies in clinical trials is hampered by our inability to distinguish between patients with FTLD-TDP and FTLD-tau. Therefore, we evaluated truncated stathmin-2 (STMN2) as a proxy of TDP-43 pathology, given the reports that TDP-43 dysfunction causes truncated STMN2 accumulation. Truncated STMN2 accumulated in human induced pluripotent stem cell–derived neurons depleted of TDP-43, but not in those with pathogenic TARDBP mutations in the absence of TDP-43 aggregation or loss of nuclear protein. In RNA-Seq analyses of human brain samples from the NYGC ALS cohort, truncated STMN2 RNA was confined to tissues and disease subtypes marked by TDP-43 inclusions. Last, we validated that truncated STMN2 RNA was elevated in the frontal cortex of a cohort of patients with FTLD-TDP but not in controls or patients with progressive supranuclear palsy, a type of FTLD-tau. Further, in patients with FTLD-TDP, we observed significant associations of truncated STMN2 RNA with phosphorylated TDP-43 levels and an earlier age of disease onset. Overall, our data uncovered truncated STMN2 as a marker for TDP- 43 dysfunction in FTD.

Find the latest version: https://jci.me/139741/pdf RESEARCH ARTICLE The Journal of Clinical Investigation

Truncated stathmin-2 is a marker of TDP-43 pathology in frontotemporal dementia

Mercedes Prudencio,1,2 Jack Humphrey,3,4 Sarah Pickles,1,2 Anna-Leigh Brown,5 Sarah E. Hill,6 Jennifer M. Kachergus,7 J. Shi,7 Michael G. Heckman,8 Matthew R. Spiegel,8 Casey Cook,1,2 Yuping Song,1 Mei Yue,1 Lillian M. Daughrity,1 Yari Carlomagno,1,2 Karen Jansen-West,1 Cristhoper Fernandez de Castro,1 Michael DeTure,1,2 Shunsuke Koga,1,2 Ying-Chih Wang,4 Prasanth Sivakumar,5 Cristian Bodo,5 Ana Candalija,9 Kevin Talbot,9 Bhuvaneish T. Selvaraj,10 Karen Burr,10 Siddharthan Chandran,10 Jia Newcombe,11 Tammaryn Lashley,12,13 Isabel Hubbard,14 Demetra Catalano,14 Duyang Kim,14 Nadia Propp,14 Samantha Fennessey,15 NYGC ALS Consortium,16 Delphine Fagegaltier,14 Hemali Phatnani,14 Maria Secrier,17 Elizabeth M.C. Fisher,5 Björn Oskarsson,18 Marka van Blitterswijk,1,2 Rosa Rademakers,1,2 Neil R. Graff-Radford,18 Bradley F. Boeve,19 David S. Knopman,19 Ronald C. Petersen,19 Keith A. Josephs,19 E. Aubrey Thompson,7 Towfique Raj,3,4 Michael Ward,6 Dennis W. Dickson,1,2 Tania F. Gendron,1,2 Pietro Fratta,5 and Leonard Petrucelli1,2 1Department of Neuroscience, Mayo Clinic, Jacksonville, Florida, USA. 2Neuroscience Graduate Program, Mayo Clinic Graduate School of Biomedical Sciences, Jacksonville, Florida, USA. 3Ronald M. Loeb Center for Alzheimer’s Disease, Nash Family Department of Neuroscience and Friedman Brain Institute, and 4Department of Genetics and Genomic Sciences, Icahn School of Medicine at Mount Sinai, New York, New York, USA. 5Department of Neuromuscular Diseases, UCL Queen Square Institute of Neurology, London, United Kingdom. 6National Institute of Neurological Disorders and Stroke (NINDS), NIH, Bethesda, Maryland, USA. 7Department of Cancer Biology, and 8Division of Biomedical Statistics and Informatics, Mayo Clinic, Jacksonville, Florida, USA. 9Nuffield Department of Clinical Neurosciences, University of Oxford, Oxford, United Kingdom. 10UK Dementia Research Institute and Euan MacDonald Centre for Motor Neurone Disease (MND) Research, The University of Edinburgh, United Kingdom. 11NeuroResource, Department of Neuroinflammation, 12Department of Neurodegenerative Disease, and 13Queen Square Brain Bank for Neurological Disorders, Department of Clinical and Movement Neuroscience, UCL Queen Square Institute of Neurology, London, United Kingdom. 14Center for Genomics of Neurodegenerative Disease, and 15New York Genome Center (NYGC), New York, New York, USA. 16The NYGC ALS Consortium is detailed in Supplemental Acknowledgments. 17University College London Genetics Institute, London, United Kingdom. 18Department of Neurology, Mayo Clinic, Jacksonville, Florida, USA. 19Department of Neurology, Mayo Clinic, Rochester, Minnesota, USA.

No treatment for frontotemporal dementia (FTD), the second most common type of early-onset dementia, is available, but therapeutics are being investigated to target the 2 main proteins associated with FTD pathological subtypes: TDP-43 (FTLD-TDP) and tau (FTLD-tau). Testing potential therapies in clinical trials is hampered by our inability to distinguish between patients with FTLD-TDP and FTLD-tau. Therefore, we evaluated truncated stathmin-2 (STMN2) as a proxy of TDP-43 pathology, given the reports that TDP-43 dysfunction causes truncated STMN2 accumulation. Truncated STMN2 accumulated in human induced pluripotent stem cell–derived neurons depleted of TDP-43, but not in those with pathogenic TARDBP mutations in the absence of TDP-43 aggregation or loss of nuclear protein. In RNA-Seq analyses of human brain samples from the NYGC ALS cohort, truncated STMN2 RNA was confined to tissues and disease subtypes marked by TDP- 43 inclusions. Last, we validated that truncated STMN2 RNA was elevated in the frontal cortex of a cohort of patients with FTLD-TDP but not in controls or patients with progressive supranuclear palsy, a type of FTLD-tau. Further, in patients with FTLD-TDP, we observed significant associations of truncated STMN2 RNA with phosphorylated TDP-43 levels and an earlier age of disease onset. Overall, our data uncovered truncated STMN2 as a marker for TDP-43 dysfunction in FTD.

Introduction 3) and, for particular genes, represses nonconserved cryptic exons Tar DNA-binding protein 43 (TDP-43) regulates multiple aspects in a species- (4) and cell-type–specific manner (5, 6). Recently, of RNA metabolism, including RNA stabilization, transcription, 2 groups independently found that TDP-43 is required for the splicing, and transport (1). TDP-43 binds thousands of RNAs (2, correct splicing and expression of stathmin-2 (STMN2) (7, 8), a microtubule-associated protein involved in axonal regeneration Related Commentary: p. 5677 (9). As shown schematically in Figure 1, when TDP-43 is present and functional, it binds to GU-rich sequences in the first intron

Authorship note: MP, JH, and SP are co–first authors and contributed equally to this work. of the STMN2 transcript, allowing normal splicing of intron 1 and Conflict of interest: RCP is a consultant for Roche Inc., Merck Inc., Biogen Inc., and repressing the usage of an alternative or cryptic polyadenylation Eisai Inc.; is a member of the data and safety monitoring board (DSMB) of Genentech; site. Reduced TDP-43 binding to STMN2 pre-mRNA, which could and has provided educational talks for GE Healthcare. occur as a result of TDP-43 nuclear depletion, leads to the inclu- Copyright: © 2020, American Society for Clinical Investigation. sion of a region within intron 1 (exon 2a) and introduces a prema- Submitted: April 30, 2020; Accepted: August 11, 2020; Published: October 19, 2020. Reference information: J Clin Invest. 2020;130(11):6080–6092. ture stop codon and polyadenylation. This causes the accumula- https://doi.org/10.1172/JCI139741. tion of a truncated variant of STMN2 that lacks exons 2 through 5.

6080 jci.org Volume 130 Number 11 November 2020 The Journal of Clinical Investigation RESEARCH ARTICLE

Figure 1. Truncated STMN2 RNA is generated with loss of nuclear TDP-43. Schematic of TDP-43–regulated STMN2 splicing. When TDP-43 is present and functional, it binds to GU-rich sequences in the first intron of the STMN2 transcript and allows normal splicing of intron 1. Reduced TDP-43 binding to STMN2 RNA, from TDP-43 aggregation or depletion from the nucleus, leads to mis-splicing of STMN2 RNA, resulting in the inclusion of a novel exon encoded within intron 1 (termed exon 2a) and containing an alternative or cryptic polyadenylation site. The production of this alternative variant of STMN2, which lacks exons 2 through 5 (referred to as truncated STMN2), is at the expense of full-length STMN2, which is reduced upon TDP-43 downregulation.

Neuronal and glial inclusions of TDP-43 are a pathologi- While approximately 50% of patients with an FTD syndrome cal hallmark of the majority (~97%) of patients with ALS and have underlying TDP-43 pathology, such pathology can, at pres- of approximately 50% of cases with the pathological subtype ent, only be definitively confirmed at autopsy. Given the efforts to of frontotemporal dementia (FTD), known as frontotempo- develop therapies specific to FTLD pathological subtypes (i.e., FTD ral lobar degeneration–TDP (FTLD-TDP) (1). ALS is a motor with TDP-43 vs. tau pathology), there is a clear need for biomarkers neuron disease (MND), whereas patients with FTD present that discriminate between them. Further, biomarkers for diagno- clinically with 1 or more of the FTD syndromes characterized sis or prognosis or that can monitor TDP-43 activity in response to by behavior, executive function, language, and motor impair- TDP-43–tailored therapies are also sorely lacking. Although TDP-43 ments (10). TDP-43 abnormalities are proposed to play a direct represents an attractive biomarker candidate to fulfill these needs, role in these diseases, since mutations in TARDBP, the gene efforts to develop sufficiently sensitive immunoassays for the detec- that encodes TDP-43, were discovered in sporadic and familial tion of TDP-43 or its pathological forms in biofluids and that reliably ALS and FTD (11, 12). It is also notable that TDP-43 morpholo- differentiate between individuals with and without TDP-43 pathol- gy and distribution are associated with clinical phenotypes and ogy have proven disappointing (27). However, nuclear clearance of neurodegeneration in key anatomical regions, such as frontal TDP-43, in the absence of TDP-43 inclusions, has been associated and temporal cortices in FTLD-TDP (13). Cells with TDP-43 with neuronal atrophy in patients with FTLD-TDP or ALS-TDP (28), pathology feature a redistribution of TDP-43 from the nucle- suggesting that early mis-splicing events due to TDP-43 loss of func- us to the cytoplasm, where TDP-43 is ubiquitinated, hyper- tion may represent a robust and early disease biomarker. phosphorylated, and cleaved into C-terminal fragments (14). We thus propose that the mis-splicing of STMN2 may provide Patients with FTLD-TDP can be further classified into differ- a novel strategy for the indirect assessment of TDP-43 dysfunc- ent pathological subtypes on the basis of the morphology and tion. To test this hypothesis, we first validated the STMN2 splic- distribution of the TDP-43 inclusions observed (15, 16). These ing event in induced pluripotent stem cell–derived (iPSC-derived) can take the form of neuronal cytoplasmic inclusions, dystro- cortical and motor neurons that were TDP-43 depleted. Then, phic neurites, neuronal intranuclear inclusions, glial inclusions, we evaluated whether the truncated STMN2 isoform accumu- perivascular inclusions, and fine neurites in the hippocampus lates in diseases and tissues characterized by TDP-43 pathology (14, 17–20). At present, 5 subtypes of FTLD-TDP (A–E) have using data sets from RNA-Seq studies of brain tissues from a large been identified (15, 20–23), with types A–C accounting for more cohort of control subjects and of patients with FTD or ALS with or than 95% of all FTLD-TDP cases (24). FTLD-TDP type A, the without TDP-43 pathology. Furthermore, we measured truncated most common subtype, is characterized by the presence of all and full-length STMN2 RNA levels in postmortem medial frontal the inclusion types mentioned above, whereas types B–E are cortex tissue from FTLD-TDP, progressive nuclear palsy (PSP), characterized by the presence of a predominant inclusion type and pathologically normal control brains. PSP was selected as the (19, 22). Furthermore, some FTLD-TDP subtypes are associat- disease control, given that TDP-43 pathology is not found in the ed with particular mutations (20, 25, 26). frontal cortex of these patients with FTLD-tau (29). We addition-

jci.org Volume 130 Number 11 November 2020 6081 RESEARCH ARTICLE The Journal of Clinical Investigation ally evaluated whether truncated STMN2 levels in the medial fron- Analyses of bulk brain tissue RNA-Seq data uncover tissue-spe- tal cortex discriminated between patients with FTLD-TDP and cific truncated STMN2 RNA accumulation in diseases characterized control subjects and assessed associations of truncated SMTN2 by TDP-43 pathology. We quantified the expression of full-length with phosphorylated TDP-43 (p–TDP-43) and clinical features. and truncated STMN2 in the NYGC ALS Consortium RNA-Seq data set across multiple neuroanatomical regions and diseases. Results As of March 2020, we were able to examine 11 tissues, with 1659 TDP-43 depletion in human iPSC–derived neurons leads to aberrant RNA-Seq samples from 439 donors matching established criteria STMN2 splicing. To confirm recent reports that correct STMN2 splic- for inclusion (see Methods). Donors were diagnosed as non-neu- ing requires TDP-43 (7, 8), we used clustered regularly interspaced rological disease controls or as having FTD, ALS, FTD with ALS short palindromic repeat inactivation (CRISPRi) to knock down TDP- (ALS-FTLD), or ALS with suspected Alzheimer’s disease (ALS- 43 expression in human iPSC–derived cortical neurons that consti- AD). Patients with FTD were classified on the basis of their neu- tutively express nuclease deactivated Cas9 (dCas9) fused to blue ropathological diagnosis: FTLD-TDP, FTLD-tau, or FTLD with fluorescent protein (BFP) and the transcriptional repressor Krüp- mutations in FUS (FTLD-FUS). ALS samples were divided into pel-associated box (KRAB) (dCas9-BFP-KRAB) (30). To achieve this, those with a SOD1 mutation (ALS-SOD1) versus those without cells were transduced with a lentivirus expressing either an sgRNA a SOD1 mutation and thus assumed to have TDP-43 patholo- against the transcriptional start site of TARDBP (TDP-43 KD), or a gy (ALS-TDP). However, pathology confirming the presence of control sgRNA predicted not to bind anywhere in the human genome TDP-43 mislocalization in patients with ALS-TDP was not sys- (controls). We confirmed silencing of TARDBP RNA by quantitative tematically reported (see Methods). reverse transcription PCR (qRT-PCR) and RNA-Seq (data set labeled For each sample, we estimated the expression of full-length “a”), which showed TDP-43–KD efficiencies of approximately 50% STMN2 using all RNA-Seq reads covering known STMN2 exons, (47.4% for qRT-PCR, 63.9% for RNA-Seq) (Figure 2A). Reduced whereas the truncated STMN2 isoform was quantified using TARDBP levels resulted in significant reductions of full-length high-confidence spliced RNA-Seq reads that unambiguously STMN2 RNA (56.7% for qRT-PCR, 65.5% for RNA-Seq) (Figure 2B) connected STMN2 exon 1 with the cryptic exon 2a (exon 1–exon and a significant increase in truncated STMN2 RNA (qRT-PCR, 32.3- 2a) (Figure 3A). We found that full-length STMN2 was highly fold increase) (Figure 2C) and RNA-Seq (Figure 2D, data set labeled expressed in both cortical and cerebellar tissues (median tran- “a”) compared with levels in neurons expressing control sgRNA. In scripts per million [TPM] in frontal cortex = 58.0 and in cere- parallel, we analyzed published RNA-Seq data from Klim et al. on bellum = 73.1) compared with expression levels in spinal cord iPSC-derived motor neurons in which TARDBP was knocked down (median TPM in lumbar spinal cord = 6.7), presumably because its (7). Knockdown of TARDBP (Supplemental Figure 1A; supplemental expression is restricted to neurons and because of the higher white material available online with this article; https://doi.org/10.1172/ matter composition in the spinal cord (Supplemental Figure 4). JCI139741DS1) was confirmed in the Klim data set (data set labeled We then estimated the expression of truncated STMN2 in the “b”) and, as expected, was accompanied by a reduction in full-length NYGC ALS cohort in disease groups across tissues. At least 2 reads STMN2 RNA levels (Supplemental Figure 1B) and an increase in trun- spanning the exon 1–exon 2a junction could be detected in 398 of cated STMN2 RNA levels (Figure 2D, data set labeled “b”). the 1659 samples in the data set (24.0%) (Supplemental Figure 5). We also investigated whether disease-causing mutations of Samples that expressed the splice junction did so at a median read TARDBP, in the absence of TDP-43 protein mislocalization or a count of 6 (0.13 reads per million) and a maximum read count of reduction of nuclear levels, are sufficient to impact STMN2 splic- 120 (2.5 reads per million) (Supplemental Figure 6). We therefore ing. We evaluated truncated STMN2 expression in RNA-Seq data chose to model truncated STMN2 expression as a binary outcome, from 2 independent studies that used iPSC-derived motor neu- with a detection threshold of 2 junction reads uniquely mapped rons from patients carrying ALS pathogenic TARDBP mutations across the intron boundary. and from control subjects (Supplemental Table 1). In contrast to Truncated STMN2 was present in disease subtypes known to neurons in which TARDBP was downregulated, we observed no have TDP-43 pathology, as it was detected in the frontal cortex of increase in truncated STMN2 in motor neurons with TARDBP 82.4% of patients with FTLD-TDP but was absent in those with mutations (Figure 2D, data sets labeled “c” and “d”). This obser- FTLD-FUS or FTLD-tau and in controls (Figure 3B). In ALS lum- vation was consistent with the lack of change in TDP-43 localiza- bar spinal cord, we detected truncated STMN2 in 62.2% of patients tion (Figure 2E), TARDBP RNA levels (Supplemental Figure 2), and with ALS with presumed TDP-43 pathology (ALS-TDP) but not in TDP-43 protein levels (Supplemental Figure 3) in TARDBP-mutant patients with ALS-SOD1 or in control individuals (Figure 3C). motor neurons. Overall, these data suggest that truncated STMN2 We observed an exquisite specificity of truncated STMN2 accumulation in iPSC-derived neurons results from a loss of TDP- expression in tissues known to have TDP-43 pathology. In patients 43 function due to TDP-43 depletion and not from splicing function with FTLD-TDP, for whom frontal cortex, temporal cortex, and changes arising from mutations in TARDBP (31). The redistribution cerebellum data sets were available, we detected truncated STMN2 of TDP-43 from the nucleus and its sequestration into cytoplasmic in frontal and temporal cortices but not in cerebellum (Figure 3D). inclusions, a pathological hallmark of FTD and ALS, with and with- In ALS-TDP tissues, which also included hippocampus, temporal out a genetic link to TARDBP, are consistent with loss-of-function cortex, motor cortex, occipital cortex, and spinal cord samples, mechanisms that would lead to STMN2 mis-splicing and an upreg- truncated STMN2 expression was most prevalent in the spinal ulation of truncated STMN2 expression and prompted us to further cord, followed by the motor cortex and frontal cortex, with little examine this in human tissue from FTD and ALS patients. to no expression detected in the occipital cortex or cerebellum,

6082 jci.org Volume 130 Number 11 November 2020 The Journal of Clinical Investigation RESEARCH ARTICLE

Figure 2. Truncated STMN2 RNA is elevated in iPSC-derived neurons with reduced levels of TDP-43. (A–C) iPSC-derived cortical neurons constitutively express- ing CRISPR inactivation machinery (data set labeled “a”) were transduced with a lentivirus expressing a sgRNA against TARDBP (TDP-43 KD) or a control sgRNA (Controls) and subjected to analysis by qRT-PCR (n = 3 TDP-43 KD, n = 3 controls) and/or RNA-Seq (n = 3 TDP-43 KD, n = 4 controls). The graphs show a decrease in TARDBP transcripts (A), a reduction of the full-length STMN2 transcript (B), and an increase in the truncated STMN2 transcript (C) upon TDP-43 depletion. (D) Detection of STMN2 cryptic exon 2a inclusion expressed as PSI by RNA-Seq in various iPSC-derived neurons: iPSC-derived cortical neurons (data set labeled “a”) described above and performed in this study; iPSC-derived motor neurons treated with TARDBP siRNA (TDP-43 KD) or control siRNA (controls) from Klim et al. (data set labeled “b”) (n = 6 TDP-43 KD, n = 11 controls) (7); iPSC-derived motor neurons from patients with TARDBP mutations (TDP-43 mut.) or controls from 2 independent groups in Edinburgh (data set labeled “c”) (n = 7 TDP-43 KD, n = 4 controls) and Oxford (data set labeled “d”) (n = 10 TDP-43 KD, n = 4 controls) (72). Data are presented as the mean ± SEM and were normalized to the control groups in A and B (set to 100%). *P < 0.05, ***P < 0.005, and ****P < 0.001, by Stu- dent’s t test (A–C) or 1-way ANOVA (D). (E) Representative images of iPSC-derived motor neurons from patients with TARDBP mutations and control subjects from a group in Oxford (data set labeled “d” in Figure 1D) showed similar nuclear-cytoplasmic distribution of TDP-43 (green) on day 30 of differentiation (DAPI in blue, choline acetyltransferase [ChAT] in white). Scale bar: 10 μm. All panels of Figure 2E are reshown in Supplemental Figure 3A. Additional data associated with this figure can also be found in Supplemental Figure 3. See also the complete unedited blots in the supplemental material.

respectively (Figure 3E and Supplemental Figure 5). As anticipated, Truncated STMN2 RNA expression is elevated in FTLD-TDP fron- patients with concurrent FTD and ALS (ALS-FTLD) had truncat- tal cortex. Given that we detected truncated STMN2 RNA in patients ed STMN2 expression in frontal and temporal cortices, as well as with FTLD-TDP but not in those with FTLD-FUS or FTLD-tau (Fig- in the motor cortex and spinal cord (Supplemental Figure 7A). In ure 3B), this aberrant splicing event may provide a means to iden- contrast, patients with ALS-AD strongly resembled patients with tify patients with FTD with marked TDP-43 pathology. To further ALS-TDP in their truncated STMN2 detection patterns, with most examine the relationship between TDP-43 and STMN2 in patients truncated STMN2 expression found in spinal cord and reduced or with FTD, we used the NanoString PlexSet platform to systemati- no expression in cortical brain regions (Supplemental Figure 7B). cally quantify truncated and full-length STMN2 RNA levels in fron- Together, these findings demonstrate that truncated STMN2 RNA tal cortex tissue from a cohort of 238 patients with FTD with immu- is detected in diseases and tissues characterized by TDP-43 pathol- nohistologically confirmed TDP-43–immunoreactive inclusions, a ogy, supporting its potential as a marker for TDP-43 dysfunction. cohort of 33 cognitively normal controls who were TDP-43 negative

jci.org Volume 130 Number 11 November 2020 6083 RESEARCH ARTICLE The Journal of Clinical Investigation

Figure 3. Truncated STMN2 RNA is detected in bulk RNA-Seq from FTLD/ALS tissues with TDP-43 pathology. (A) The NYGC ALS Consortium RNA-Seq data set was analyzed for the presence of truncated STMN2 transcripts, which were identified by RNA-Seq reads spanning the exon 1–exon 2a splice junction. (B–E) Bar graphs represent the proportion of individuals (percentage) with at least 2 reads spanning the exon 1–exon 2a junction. The number of individuals for the samples of the indicated tissues/diseases is indicated. (B) Truncated STMN2 was detected in the frontal cortex of patients with FTLD- TDP but not in control subjects or patients with FTLD-FUS or FTLD-tau. (C) In the lumbar spinal cord of patients with ALS, truncated STMN2 RNA was only detected in ALS-TDP samples but not in ALS-SOD1 or control samples. (D) Among FTLD-TDP tissues, truncated STMN2 was only detected in those known to have TDP-43 pathology (frontal and temporal cortex), whereas (E) within ALS-TDP tissues, truncated STMN2 RNA was seen in motor cortex and spinal cord tissues but not in regions without TDP-43 pathology.

(henceforth referred to as controls), and a cohort of 41 patients with that truncated STMN2 RNA levels were significantly elevated in all PSP (Table 1). The latter cohort of patients were characterized by patients with FTLD-TDP compared with levels in the control sub- tau pathology and neurodegeneration in the frontal cortex. jects without disease or those with PSP disease in both unadjusted As mentioned, FTLD-TDP can be further classified on the analysis (P < 0.001) (Figure 4A, Table 2, and Supplemental Table basis of morphology and distribution of TDP-43 inclusions. As 2) and analysis adjusted for age at death, sex, and RNA integrity such, our cohort of patients with FTLD-TDP were classified into number (RIN) (P = 0.009 vs. controls, P < 0.001 vs. patients with different TDP-43 pathological subtypes: 117 patients were type A PSP) (Figure 4A, Table 2, and Supplemental Table 2). We also (27 and 34 patients carried a mutation in C9orf72 or GRN, respec- noted that the levels of truncated STMN2 in individuals with PSP tively), 66 were type B (21 and 1 carried a mutation in C9orf72 or were significantly lower than those in control subjects (P = 0.028, GRN, respectively), 43 were type C (2 and 1 carried a mutation in adjusted analyses for age, sex, and RIN), possibly because of the C9orf72 or GRN, respectively), and 2 were type D (both patients neuronal loss in the patients with PSP. had VCP mutants) (Table 1). In addition to conducting our anal- As in all patients with FTLD-TDP combined, truncated STMN2 yses of all FTLD-TDP cases combined, we also evaluated wheth- RNA levels in the frontal cortex were increased in FTLD-TDP for er truncated STMN2 accumulation was preferentially associated each TDP-43 type (A–D) compared with levels in patients with with a particular TDP-43 subtype. In the frontal cortex, we found PSP in both unadjusted and adjusted analyses (Table 2 and Sup-

6084 jci.org Volume 130 Number 11 November 2020 The Journal of Clinical Investigation RESEARCH ARTICLE

Table 1. Patient characteristics in the postmortem cohort

Variable Controls PSP All FTLD-TDP FTLD-TDP type A FTLD-TDP type B FTLD-TDP type C FTLD-TDP type D (n = 33) (n = 41) (n = 238) (n = 117) (n = 66) (n = 43) (n = 2) Age at death (yr) 82 (55, 99) 69 (52, 86) 72 (49, 100) 76 (50, 100) 67 (49, 98) 74 (60, 84) 63 (61, 64) Sex Male 20 (60.6%) 13 (51.2%) 130 (54.6%) 63 (53.8%) 37 (56.1%) 23 (53.5%) 1 (50.0%) Female 13 (39.4%) 20 (48.8%) 108 (45.4%) 54 (46.2%) 29 (43.9%) 20 (46.5%) 1 (50.0%) Survival after onset (yr) NA 7.0 (4.0, 14.0) 7.0 (0.6, 25.0) 7.0 (1.0, 25.0) 4.0 (0.6, 20.0) 10.6 (4.0, 25.0) 8.5 (7.0, 10.0) Age at onset (yr) NA 61.5 (46.0, 79.5) 64.5 (40.0, 85.8) 67.8 (40.0, 85.8) 61.4 (44.0, 85.7) 62.7 (46.0, 76.0) 54.1 (54.0, 54.2) RIN 9.6 (7.1, 10.0) 9.5 (7.1, 10.0) 9.5 (7.0, 10.0) 9.4 (7.0, 10.0) 9.6 (7.0, 10.0) 9.5 (7.8, 10.0) 9.9 (9.8, 10.0) The sample median (minimum, maximum) is given for continuous variables. Information was unavailable regarding TDP-43 subtype (all FTLD-TDP: n = 10), age at death (all FTLD-TDP: n = 1), p–TDP-43 burden in the frontal cortex (all FTLD-TDP: n = 1), survival after onset (PSP: n = 4; all FTLD-TDP: n = 20), and age at onset (PSP: n = 4; all FTLD-TDP: n = 20). Among the patients with FTLD-TDP, 50 had C9orf72 mutations (n = 27 TDP-43 type A, n = 21 TDP-43 type B, n = 2 TDP-43 type C), 36 had GRN mutations (n = 34 TDP-43 type A, n = 1 TDP-43 type B, n = 1 TDP-43 type C), and 2 had VCP mutations (TDP-43 type D).

plemental Table 2). Compared with controls, truncated STMN2 (FTLD-tau, FTLD-FUS) or ALS (ALS-SOD1) (Figure 3) suggests levels were significantly increased in patients with FTLD-TDP that truncated STMN2 production is caused by a loss of TDP-43 types A, B, and C, but not in those with type D, in unadjusted anal- nuclear function resulting from its mislocalization to the cyto- yses (Table 2 and Supplemental Table 2), although the levels only plasm and/or its sequestration into inclusions — hallmark features in those with type A FTLD-TDP remained significant in analyses of FTLD-TDP. To evaluate this possibility further, we examined adjusted for age, sex, and RIN (Table 2 and Supplemental Table 2). whether full-length or truncated STMN2 RNA levels in the fron- We observed that truncated STMN2 levels in the frontal cortex tal cortex correlated with p–TDP-43 accumulation. To this end, discriminated between the patients with FTLD-TDP and the PSP we prepared urea-soluble protein lysates from the same tissues control subjects, with an area under the receiver operating charac- used to measure STMN2 isoforms and quantified p–TDP-43 in teristic curve (AUC) of 0.76 (95% CI = 0.69–0.84) (Table 2). Frontal these lysates using a previously described immunoassay (32–34). cortex truncated STMN2 also discriminated between FTLD-TDP Although the levels of full-length STMN2 did not associate with patients and cognitively normal controls with a more modest AUC p–TDP-43 levels (β: 0.02, P = 0.53) (Supplemental Table 4), higher of 0.67 (95% CI = 0.57–0.78) (Supplemental Table 2). Further- p–TDP-43 levels were associated with higher truncated STMN2 RN more, truncated STMN2 levels similarly discriminated between A levels (Figure 4C) in both unadjusted analyses (β: 0.20, controls (cognitively normal or PSP disease controls) and patients P < 0.001) (Table 3) and in a multivariable analysis adjusted for with FTLD-TDP of all TDP-43 types, most prominently for types age, sex, and TDP-43 subtype (β: 0.19, P < 0.001) (Table 3). Tak- A, C, and D (Table 2 and Supplemental Table 2). In contrast, trun- en together, our data suggest that truncated STMN2 RNA, but not cated STMN2 levels did not effectively distinguish patients with full-length STMN2, accumulates in brain regions marked by TDP- PSP from control subjects (AUC: 0.60, 95% CI = 0.47–0.73). 43 pathology and associates with p–TDP-43 burden. The increase in truncated STMN2 levels in FTLD-TDP fron- Truncated STMN2 RNA levels associate with an earlier age of tal cortex was paired with a decrease in full-length STMN2 levels FTLD-TDP onset. We additionally examined whether truncated or (P < 0.001) (Figure 4B and Supplemental Table 3). Of interest, full- full-length STMN2 levels in the frontal cortex associate with clinical length STMN2 expression was also decreased in patients with PSP characteristics of individuals with FTLD-TDP. We found a significant compared with expression levels in controls (P < 0.001) (Figure inverse correlation between truncated STMN2 RNA levels and age at 4B and Supplemental Table 3), probably because of neurodegen- disease onset (Figure 4D) in analyses, adjusting for sex, age at death, eration in this region. Full-length STMN2 levels also distinguished and TDP-43 type and after correcting for multiple testing (P = 0.005) control subjects from patients with FTLD-TDP (AUC: 0.74) (Table 3). However, survival after onset did not associate with trun- and PSP (AUC: 0.83), but were not as effective in distinguishing cated STMN2 RNA levels, after correcting for age at onset, sex, and patients with FTLD-TDP from those with PSP (AUC: 0.60, 95% TDP-43 subtype (P = 0.32) (Table 3). Of interest, we observed a nom- CI = 0.51-0.68; Supplemental Table 3). inal but not statistically significant association with lower truncated Taken together, our data show that truncated STMN2 RNA STMN2 levels in males (β: –0.35, P = 0.034 in adjusted analyses for levels were significantly elevated in patients with FTLD-TDP age and TDP-43 type) (Supplemental Table 4). In contrast, full-length compared with levels in controls and may be used to distinguish STMN2 levels failed to associate with clinical features of FTLD-TDP, FTD patients with TDP-43 pathology from both cognitively nor- including age of onset and survival (Supplemental Table 4). mal individuals and disease controls. Truncated STMN2 RNA associates with p–TDP-43 burden in the Discussion frontal cortex of patients with FTLD-TDP. That truncated STMN2 FTLD, the neuropathological substrate of FTD, is the second levels were elevated in the frontal cortex of patients with FTLD- most common cause of early-onset dementia (35). Although no TDP but not in those with PSP (Figure 4A) or other forms of FTLD effective treatment is available, several therapeutic approaches

jci.org Volume 130 Number 11 November 2020 6085 RESEARCH ARTICLE The Journal of Clinical Investigation

Figure 4. Truncated STMN2 is elevated in FTLD-TDP and associates with higher p–TDP-43 burden and earlier age of FTD onset. (A and B) Truncated and full- length STMN2 levels were measured in RNA extracted from frontal cortex of control subjects and FTLD-TDP and PSP patients, using the NanoString PlexSet platform. (A) A significant accumulation of truncated STMN2 RNA was detected in patients with FTLD-TDP but not in controls (see also Supplemental Table 2) or patients with PSP (see also Table 2). (B) Full-length STMN2 levels were significantly decreased in FTLD-TDP and PSP patients (see also Supplemental Table 3). Data are presented as the median with a 95% CI, and P values were determined using linear regression models that were adjusted for age at death, sex, and RIN. *P < 0.05, ***P < 0.005, ****P < 0.001. (C and D) In our FTLD-TDP cohort (n = 238), significant correlations were observed between truncated STMN2 RNA levels and (C) a higher burden of p–TDP-43 and (D) an earlier age of disease onset. See Table 3 for the correlation coefficients.

are being investigated to target TDP-43 and tau, the 2 proteins neurons with pathogenic TARDBP mutations where no change most commonly associated with the major pathological FTLD in TDP-43 levels, localization, or aggregates was detected, sup- subtypes. Their testing in clinical trials, however, faces a consid- porting the fact that loss of TDP-43 nuclear function, and not erable barrier: our inability to determine the underlying protein splicing changes induced by pathogenic mutations, were driv- pathology in living patients with FTD. Although the pathomech- ing this event. We were unable to assess postmortem tissue anisms driven by TDP-43 and tau differ, the clinical symptoms of from patients carrying TARDBP mutations, which are rare in FTD patients with TDP-43 or tau pathology overlap. Ultimately, the FTD and ALS populations, as a group, but given that these this hinders the selection of patients with sporadic FTD who are patients develop florid TDP-43 pathology, we expected trun- suitable for a particular clinical trial. Since there is a clear need cated STMN2 to be present at levels similar to those in patients for biomarkers to identify FTD patients with TDP-43 pathology, with sporadic ALS or FTLD-TDP. Additionally, we found that we evaluated whether truncated STMN2 could serve as a proxy truncated STMN2 RNA accumulated in the frontal cortex of for TDP-43 pathology. Using RNA-Seq and RNA quantification patients with FTLD-TDP but not in control subjects or patients techniques, we show that truncated STMN2 RNA levels were with PSP and was significantly correlated with p–TDP-43 levels elevated in human iPSC–derived neurons depleted of TDP-43 and an earlier age of disease onset. Taken together, our data and in bulk tissue from FTD and ALS patients with TDP-43 support the notion that truncated STMN2 production results pathology. We did not detect truncated STMN2 in iPSC-derived from TDP-43 loss of function.

6086 jci.org Volume 130 Number 11 November 2020 The Journal of Clinical Investigation RESEARCH ARTICLE

Table 2. Comparisons of truncated STMN2 RNA in frontal cortex between patients with FTLD-TDP and patients with PSP.

Unadjusted analysis Adjusting for age at death, sex, and RIN Group n Median (minimum, maximum) Regression coefficient P value Regression coefficient P value AUC (95% CI) levels of truncated STMN2 RNA (95% CI) (95% CI) vs. PSP cases PSP 41 217 (22, 880) 0.00 (reference) NA 0.00 (reference) NA NA All FTLD-TDP 238 560 (8, 5210) 1.38 (0.96, 1.80) < 0.001 1.36 (0.95, 1.77) < 0.001 0.76 (0.69, 0.84) FTLD-TDP type A 117 595 (8, 3495) 1.47 (0.98, 1.95) < 0.001 1.37 (0.87, 1.86) < 0.001 0.78 (0.70, 0.86) FTLD-TDP type B 66 494 (62, 5210) 1.28 (0.77, 1.79) < 0.001 1.29 (0.79, 1.78) < 0.001 0.74 (0.65, 0.84) FTLD-TDP type C 43 575 (106, 2071) 1.34 (0.79, 1.89) < 0.001 1.26 (0.73, 1.78) < 0.001 0.77 (0.66, 0.87) FTLD-TDP type D 2 1035 (1008, 1062) 2.40 (0.25, 4.55) 0.029 3.07 (1.07, 5.07) 0.004 1.00 (1.00, 1.00) Regression coefficients, 95% CIs, and P values were determined using linear regression models, where the levels of truncated STMN2 RNA in the frontal cortex were considered on the base 2 logarithmic scale. Regression coefficients were interpreted as the difference in mean levels of truncated STMN2 RNA in the frontal cortex (on the base 2 logarithmic scale) between the given group of patients with disease and patients with PSP. P values of 0.010 or less were considered statistically significant after applying a Bonferroni’s correction for multiple testing.

As mentioned above, 2 independent reports recently revealed logical changes in TDP-43. In line with this, we observed a sig- that TDP-43 regulates STMN2 expression and that TDP-43 deple- nificant association of truncated STMN2 RNA with p–TDP-43 tion produces a truncated, nonfunctional STMN2 variant (7, 8). protein levels (P < 0.001, Table 3), which remained statistical- Increased truncated STMN2 RNA levels were detected in spinal ly significant after controlling for multiple variables (age, sex, cord tissue of 17 patients with ALS but not in that of 6 controls, and and TDP-43 subtype; P < 0.001, Table 3). However, it is possible ISH studies found truncated STMN2 RNA in lower and upper motor that p–TDP-43 levels alone may not be a perfect corollary for neurons in 5 patients with ALS (8). We thus evaluated truncated nuclear TDP-43 depletion. Of interest, the levels of truncated STMN2 as a proxy TDP-43 activity marker for FTD using postmor- STMN2 RNA did not differ among different TDP-43 subtypes, tem RNA-Seq data across 11 tissues available from a large cohort suggesting that truncated STMN2 levels are elevated in all types of FTD, ALS, and FTLD-ALS patients. Notably, our RNA-Seq anal- of TDP-43 pathology. yses detected the STMN2 cryptic exon that gives rise to truncated Importantly, higher truncated STMN2 RNA levels in the fron- STMN2 only in FTD and ALS patients with TDP-43 pathology. In tal cortex at death were significantly associated with an earlier age addition, the presence of abnormal STMN2 was highly specific to at disease onset (P = 0.005, Table 3), but we found no association brain regions characterized by TDP-43 pathology. with survival after disease onset. Given that truncated STMN2 In addition to our analyses of RNA-Seq data, we directly RNA serves as a proxy for TDP-43 dysfunction, this earlier age measured truncated STMN2 RNA in postmortem frontal cortex of onset in patients with higher postmortem truncated STMN2 tissue from patients with FTLD-TDP, which again revealed an RNA levels could point to more severe defects in RNA metabo- aberrant accumulation of truncated STMN2 in comparison with lism resulting from aberrant TDP-43 function, including loss of controls and individuals with PSP. In fact, we found that truncated STMN2 function. STMN2 RNA levels distinguished between patients with FTLD- Strengths of our study include the use of samples from a clini- TDP and those with PSP, with an area under the ROC curve of cally well-characterized, immunohistologically confirmed cohort 0.76. As anticipated, truncated STMN2 accumulation in patients of 238 patients with FTLD-TDP from which high-integrity RNA with FTLD-TDP was accompanied by a decrease in full-length was extracted from the frontal cortex, and the use of quantitative STMN2, as was observed in cell culture models (7, 8). Full-length measurements of truncated STMN2 RNA and p–TDP-43 protein. STMN2 was also decreased in patients with PSP, which were con- Moreover, we evaluated multiple neuroanatomical regions in bulk firmed to lack TDP-43 pathology. Given that STMN2 is expressed tissue RNA-Seq from a large data set that included 1659 FTD and by neurons and its expression is abundant in the frontal cortex, the ALS patients with and without TDP-43 pathology. Further, our decrease in full-length STMN2 in patients with PSP likely reflects study also used 3 different approaches (RNA-Seq, NanoString, neurodegenerative neuronal loss. Indeed, a recent report found and qRT-PCR, the latter of which was also performed to validate that full-length STMN2 RNA levels were also decreased in the truncated STMN2 in patients with FTLD-TDP, as shown in Sup- brains of patients with Parkinson’s disease (36), possibly arguing plemental Figure 9) that have different sensitivity and specificity for the use of STMN2 levels as a neurodegenerative disease bio- linked to the methodology but that all detect significant differ- marker. It is also possible that, aside from TDP-43, various other ences in truncated STMN2, highlighting the reproducibility of the factors regulate full-length STMN2 expression. This could explain finding. Our study also has some limitations. While we specifically why full-length STMN2 levels did not correlate with p–TDP-43 designed our study to evaluate truncated STMN2 as a marker of burden or with clinical features. TDP-43 pathology — information important to establish its value The accumulation of truncated STMN2 RNA in the frontal as a candidate biomarker — we did not measure the translation cortex of patients with FTLD-TDP but not of patients with PSP product from this truncated variant because of a lack of com- indicates that truncated STMN2 production is caused by patho- mercially available antibodies. At present, it is unclear whether

jci.org Volume 130 Number 11 November 2020 6087 RESEARCH ARTICLE The Journal of Clinical Investigation

Table 3. Associations of truncated STMN2 with p–TDP-43 and clinical characteristics in FTLD-TDP frontal cortex

Association with frontal cortex truncated STMN2 RNA Unadjusted analysis Multivariable analysis Variable Regression coefficient P value Regression coefficient P value Multivariable model adjustments (95% CI) (95% CI) p–TDP-43 (doubling) 0.20 (0.12, 0.28) < 0.001 0.19 (0.09, 0.28) < 0.001 Age at death, sex, TDP-43 subtype TDP-43 subtype Overall test of difference: P = 0.48 Overall test of difference: P = 0.33 Age at death and sex FTLD-TDP type A 0.00 (reference) NA 0.00 (reference) NA FTLD-TDP type B –0.19 (–0.57, 0.19) 0.33 –0.31 (–0.72, 0.10) 0.14 FTLD-TDP type C –0.13 (–0.56, 0.31) 0.57 –0.16 (–0.60, 0.27) 0.46 FTLD-TDP type D 0.93 (–0.91, 2.77) 0.32 0.68 (–1.17, 2.53) 0.47 Age at onset (10-yr increase) –0.23 (–0.41, –0.04) 0.015 –0.29 (–0.49, –0.09) 0.005 Sex and TDP-43 subtype Survival after onset (5-yr increase) 0.12 (–0.06, 0.30) 0.19 0.10 (–0.09, 0.29) 0.32 Age at onset, sex, and TDP-43 subtype Sex Male –0.37 (–0.68, –0.06) 0.019 –0.35 (–0.67, –0.03) 0.034 Age at death and TDP-43 subtype Female 0.00 (reference) NA 0.00 (reference) NA Regression coefficients, 95% CIs, and P values were determined using linear regression models, where frontal cortex truncated STMN2 RNA was considered on the base 2 logarithmic scale. Regression coefficients were interpreted as the change in mean frontal cortex truncated STMN2 RNA (on the base 2 logarithmic scale) corresponding to the presence of the indicated characteristic (categorical variables) or the increase shown in parentheses (continuous variables). P values of 0.010 or less were considered statistically significant after applying a Bonferroni’s correction for multiple testing.

the predicted peptide of only 16 amino acids is translated from AD develop TDP-43 inclusions, and, compared with AD patients truncated STMN2 RNA. Nevertheless, our findings indicate that without TDP-43 pathology, they are cognitively and functionally the development of novel tools (antibodies, immunoassays) for worse, suggesting that abnormal TDP-43 causes a modified AD the detection of truncated STMN2 protein and additional stud- phenotype (37–46). Our work also highlights how future means to ies to evaluate the role of truncated STMN2 RNA and protein in detect truncated STMN2 RNA or protein in biofluids may provide disease pathogenesis and as biomarkers are well merited. Criti- important information on disease activity and therefore be useful cal next steps to hasten the translation of this finding to the clinic in assessing drug efficacy in clinical trials. Thus, these endeavors include determining whether truncated STMN2 RNA or protein are expected to increase the likelihood of success in developing is detected in the biofluids of patients with FTD (e.g., spinal flu- effective therapeutics and improving the care of patients with id and blood) and, if so, whether they discriminate between FTD TDP-43 proteinopathies. subtypes (i.e., with or without TDP-43). Although expression of truncated STMN2 in affected postmortem tissues can distinguish Methods FTLD-TDP from PSP cases, which has important implications, as RNA extraction, complementary DNA synthesis, and qRT-PCR in iPSC-de- it supports truncated STMN2 as a viable proxy for the presence of rived neurons. iPSC-derived neurons were harvested, and cell pellets were TDP-43 pathology, the ability to discriminate between FTLD-TDP lysed in TRI reagent (Zymo Research). Then, total RNA was extracted fol- and PSP patients was only modest (AUC: 0.76). RNA degradation lowing the manufacturer’s instructions using the Direct-zol RNA Miniprep invariably occurs in postmortem brain tissues, possibly contrib- Kit (Zymo Research), and 850 ng RNA was used for RT-PCR. RT-PCR of uting to this modest AUC and potentially pointing toward the RNA samples was performed using the High Capacity cDNA Transcrip- possibility of detecting truncated STMN2 in CSF or blood, which tion Kit with random primers (Applied Biosystems), and qRT-PCR was are typically frozen immediately after collection, to discriminate performed using SYBR GreenER qPCR SuperMix (Invitrogen, Thermo between patient groups. Nonetheless, sensitive methods will be Fisher Scientific). Samples were run in triplicate, and qRT-PCRs were run needed to detect truncated STMN2 RNA and/or protein in human on a Prism 7900HT Fast Real-Time system (Applied Biosystems). Relative biofluids. Additionally, determining at what point during the quantification was determined using the ΔΔCt method and normalized to course of disease truncated STMN2 levels begin to be upregulat- the endogenous controls RPLP0 and GAPDH. The following primers and ed will have to be evaluated and will probably require longitudinal their sequences were used: STMN2 forward: 5′-AGCTGTCCATGCT- collections of CSF and blood first from patients with ALS, who are GTCACTG-3′, STMN2 reverse: 5′-GGTGGCTTCAAGATCAGCTC-3′; almost certain to develop TDP-43 pathology, and then from indi- truncated STMN2 forward: 5′-GGACTCGGCAGAAGACCTTC-3′, trun- viduals with FTD or FTD/ALS harboring genetic variants associ- cated STMN2 reverse: 5′-GCAGGCTGTCTGTCTCTCTC-3′; TARDBP ated with TDP-43 dysfunction. forward: 5′-AATTCTGCATGCCCCAGA-3′, TARDBP reverse: 5′-GAAG- If, as anticipated, truncated STMN2 RNA or protein is detect- CATCTGTCTCATCCATTTT-3′; GAPDH forward: 5′-GTTCGACAGT- ed in the biofluids of FTD patients with TDP-43 pathology, it may CAGCCGCATC-3′, GAPDH reverse: 5′-GGAATTTGCCATGGGTG- have the potential to similarly identify those patients with AD GA-3′; and RPLP0 forward: 5′-TCTACAACCCTGAAGTGCTTGAT-3′, who have TDP-43 pathology. Indeed, up to 67% of patients with RPLP0 reverse: 5′-CAATCTGCAGACAGACACTGG-3′.

6088 jci.org Volume 130 Number 11 November 2020 The Journal of Clinical Investigation RESEARCH ARTICLE

RNA-Seq analyses from iPSC-derived neurons. Data from Klim et RNA-Seq samples were subjected to quality control modeled on al. (7) were downloaded from the NCBI’s Gene Expression Omnibus the criteria of the Genotype Tissue Expression Consortium (64). Any (GEO) database (GEO GSE121569). All iPSC cells lines were analyzed sample failing 1 of the following sequencing metric thresholds was with the same analysis pipeline. Raw FASTQ files were trimmed using removed: a unique alignment rate of less than 90%, ribosomal bases fastp (47) and aligned to the hg38 build (GRCh38.primary_assembly) of greater than 10%, a mismatch rate of greater than 1%, a duplica- of the human reference genome using STAR (2.7.2a) (48). Gene abun- tion rate of greater than 0.5%, intergenic bases of less than 10.5%, and dance for TARDBP and STMN2 was calculated with featureCounts (49) ribosomal bases of greater than 0.1%. Sex was verified using XIST and using gene models from GENCODE GTF, version 31, and converted to UTY expression. Tissue identity was confirmed using the expression reads per kilobase per million (RPKM) using edgeR (50). The relative of the cerebellar marker CBNL1 and the cortical marker NRGN. Spinal abundance of truncated STMN2 as a proportion of total STMN2 was cord samples were confirmed by expression of the oligodendrocyte expressed in terms of percent spliced in (PSI) (51). This was calculated marker MBOP. Following quality control, 1905 samples of 1924 were by parsing the splice junction tables from STAR and taken as the per- retained for analysis. When comparing subgroups and tissues, only centage of all uniquely mapped STMN2 junction reads from exon 1 to tissues with at least 50 samples and disease subgroups with at least 4 exon 2a over the sum of all junction reads coming from exon 1. individuals were kept for analysis. This resulted in 1659 samples from NYGC ALS Consortium cohort. Patients with FTD were classi- 11 tissues and 439 donors. fied according to a pathologist’s diagnosis of FTD with TDP-43 Expression of the full-length STMN2 isoform was estimated inclusions (FTLD-TDP), tau inclusions (FTLD-tau), or FUS inclu- using RSEM, summing the estimates of the 3 annotated isoforms sions (FTLD-FUS). ALS samples were divided into the following in GENCODE, version 30, into a gene-level estimate of expression subcategories using the available Consortium metadata: ALS with normalized in terms of TPM. Given the large size of the NYGC ALS or without reported SOD1 mutations (ALS-TDP and ALS-SOD1); cohort, RNA-Seq data were produced on 2 different sequencing plat- ALS with frontotemporal dementia (ALS-FTLD); and ALS with forms. Exploratory analyses showed a substantial batch effect in gene AD (ALS-AD). All non-SOD1 ALS samples were grouped as “ALS- expression due to the 2 sequencing platforms. We therefore split the TDP” in this work for simplicity, although reporting of postmortem data set by sequencing platform, along with tissue and disease status. TDP-43 inclusions was not systematic and therefore not integrated A comparison of ALS-TDP and non-neurological controls revealed into the metadata. Confirmed TDP-43 pathology postmortem was that STMN2 had an inconsistent pattern of differential expression, reported for all FTLD-TDP samples. with no tissue showing a marked difference in either platform (Sup- NYGC ALS Consortium RNA-Seq library preparation from bulk tis- plemental Figure 4A). In addition, RNA degradation is known to sue RNA. The Consortium’s sample processing has been, in part, pre- affect the estimation of transcript abundance (65, 66), and we rea- viously described (52). In brief, RNA was extracted from flash-frozen soned that systematic differences in RNA quality among samples postmortem tissue using TRIzol (Thermo Fisher Scientific) chloro- sequenced on different platforms and between disease and control form, followed by column purification (RNeasy Minikit, QIAGEN). groups or tissues could confound our estimation of STMN2 expres- RIN was assessed on a Bioanalyzer (Agilent Technologies). RNA- sion. We compared the RIN values for each sample between the ALS- Seq libraries were prepared from 500 ng total RNA using the KAPA TDP and control groups and found significant differences in RIN Stranded RNA-Seq Kit with RiboErase (KAPA Biosystems) for rRNA values in some comparisons (Supplemental Figure 4B). The small depletion and Illumina-compatible indexes (NEXTflex RNA-Seq Bar- number of controls available limited further comparisons in some tis- codes, NOVA-512915, PerkinElmer, and IDT for Illumina TruSeq UD sues. We then correlated the RIN with STMN2 expression and found Indexes, 20022370). Pooled libraries (average insert size: 375 bp) a positive correlation in multiple tissues and on both sequencing plat- passing the quality criteria were sequenced either on an Illumina HiS- forms (Supplemental Figure 4C). eq 2500 (125 bp paired end) or an Illumina NovaSeq (100 bp paired Quantifying STMN2 splicing from RNA-Seq data. For downstream end). Samples were subjected to extensive sequencing and RNA-Seq analysis of STMN2 splice junction reads, a bioinformatics pipeline was quality control metrics at the NYGC that are described below. Nota- created using Snakemake (67, 68). Uniquely mapped reads within the bly, a set of more than 250 markers was used to confirm tissue, neu- STMN2 locus were extracted from each sample using SAMtools. Any roanatomical regions, and sex in the RNA-Seq data. Samples passing read marked as a PCR duplicate by Picard Tools was discarded. Splice these metrics are available for distribution. The samples had a median junction reads were then extracted with RegTools (69) using a mini- sequencing depth of 42 million read pairs, with a range between 16 mum of 8 bp as an anchor on each side of the junction and a maximum and 167 million read pairs. intron size of 500 kb. Junctions from each sample were then clustered Analyses of bulk tissue RNA-Seq data. Samples were uniformly together using LeafCutter (70) with relaxed junction filtering (minimum processed using RAPiD-nf, an efficient RNA-Seq processing pipeline total reads per junction = 30, minimum fraction of total cluster reads = implemented in the NextFlow framework (53). Following adapter 0.0001). This produced a matrix of junction counts across all samples. trimming with Trimmomatic (version 0.36) (54), all samples were Using coordinates from the Klim (7) and Melamed (8) publications, the aligned to the hg38 build (GRCh38.primary_assembly) of the human TDP-43–associated junction of the novel truncated STMN2 isoform was reference genome using STAR (2.7.2a) (48), with indexes created from extracted from the matrix. A detection threshold of 2 or more unique- GENCODE, version 30 (55). Gene expression was quantified using ly mapped reads was applied. We found no association between trun- RSEM (1.3.1) (56). Quality control was performed using SAMtools (57) cated STMN2 detection and RIN, sequencing platform, or total library and Picard Tools (58), and the results were collated using MultiQC size (Supplemental Figure 8, A–C). The relationship between expression (59). Exploratory plots were created in R, version 3.6.0 (60), using levels of full-length and truncated STMN2 was complex, with no overall dplyr, tidyr, stringr, ggplot2, patchwork, and ggpubr packages (61–63). detection of association across tissues (Supplemental Figure 8D).

jci.org Volume 130 Number 11 November 2020 6089 RESEARCH ARTICLE The Journal of Clinical Investigation

RNA extraction and NanoString analyses in postmortem tissues. performed for the separate FTLD-TDP versus control and FTLD- Using the RNAeasy Plus Mini Kit (QIAGEN), RNA was extracted TDP versus PSP analyses, after which P values of less than 0.01 were from frozen human postmortem frontal cortex tissues (33). RNA considered statistically significant. with a RIN of 7.0 or higher, as determined using an Agilent 2100 In a secondary analysis, we compared truncated STMN2 RNA lev- bioanalyzer (Agilent Technologies), was used to measure transcript els between controls and patients with PSP and made pairwise com- levels using the NanoString PlexSet platform (250 ng human brain parisons of full-length STMN2 RNA and p–TDP-43 (both on the base RNA, according to the manufacturer’s instructions). Transcript lev- 2 logarithmic scale) between controls, patients with PSP, and patients el abundance was determined using nSolver 4.0 software (Nano­ with FTLD-TDP using the aforementioned linear regression analyses. String Technologies) and normalized to HTRP1, a housekeeping In the subgroup of 238 patients with disease, associations of 5 dif- gene that was unaltered across disease subtypes in our data set ferent variables (p–TDP-43 levels, TDP-43 subtype, age at onset, sur- and previously used to normalize human brain transcripts (71). The vival after onset, and sex) with both truncated and full-length STMN2 sequences of the NanoString probes used in this study are included RNA levels were evaluated using single-variable and multivariable lin- in Supplemental Table 5. ear regression models, where truncated and full-length STMN2 RNA p–TDP-43 immunoassay. The p–TDP-43 immunoassay was per- levels in the frontal cortex were assessed on the base 2 logarithmic formed as previously described (33). Briefly, approximately 50 mg scale, and multivariable models were adjusted for predefined, poten- postmortem tissue from the frontal cortex of patients or unaffect- tially confounding variables. The multivariable model assessing the ed individuals were homogenized in cold RIPA buffer (25 mM Tris- p–TDP-43 response in frontal cortex was adjusted for age at death, sex, HCl, pH 7.6, 150 mM NaCl, 1% sodium deoxycholate, 1% Nonidet and TDP-43 subtype. The multivariable model assessing TDP-43 sub- P-40, 0.1% sodium dodecyl sulfate, and protease and phosphatase type was adjusted for sex and age at death. The multivariable model inhibitors) and sonicated on ice. Homogenates were centrifuged at assessing age at onset was adjusted for sex and TDP-43 subtype. The 100,000g for 30 minutes at 4°C. The supernatant was collected, and multivariable model assessing survival after onset was adjusted for the pellet was resuspended in RIPA buffer, sonicated and centrifuged age at onset, sex, and TDP-43 subtype. Finally, the multivariable mod- again to prevent carryover of soluble material. The RIPA-insoluble el assessing sex was adjusted for age at death, TDP-43 subtype, and pellet was extracted using 7 M urea buffer and then sonicated and genotype. Regression coefficients and 95% CIs were estimated and centrifuged at 100,000g for 30 minutes at 22°C. The protein concen- are interpreted as the change in the mean level of truncated or full- tration of the urea-soluble supernatant was determined by Bradford length STMN2 RNA in the frontal cortex (on the base 2 logarithmic assay. p–TDP-43 levels were determined using a sandwich Meso Scale scale) corresponding to the presence of the given characteristic (cat- Discovery (MSD) immunoassay. The capture antibody was a mouse egorical variables) or a specified increase (continuous variables). The monoclonal antibody that detects TDP-43 phosphorylated at serines p–TDP-43 response in the frontal cortex was examined on the base 2 409/410 (1:500, no. CAC-TIP-PTD-M01, Cosmo Bio USA), and the logarithmic scale in linear regression analysis because of its skewed detection antibody was a sulfo-tagged rabbit polyclonal C-terminal distribution. In order to adjust for the 5 different variables that were TDP-43 antibody (2 μg/mL, 12892-1-AP, Proteintech). Urea-solu- assessed for association with the separate truncated and full-length ble fractions were diluted in TBS to 35 μg protein per well and test- STMN2 RNA outcomes, we used Bonferroni’s correction for multiple ed in duplicate wells. Response values corresponding to the intensity testing, after which P values of less than 0.01 were considered statis- of emitted light upon electrochemical stimulation of the assay plate tically significant. All statistical tests were 2 sided. All statistical anal- using the MSD QUICKPLEX SQ120 were acquired. yses described above were performed using R Statistical Software, Statistics. Comparisons of truncated STMN2 RNA in the frontal version 3.6.2 (R Foundation for Statistical Computing). cortex between the separate non–FTLD-TDP groups of controls and Statistical analyses for iPSC studies and qRT-PCR of RNA extract- patients with PSP and various groups of patients with FTLD-TDP ed from human tissues were performed using GraphPad Prism 8 (all FTLD-TDP patients as well as 4 different subgroups according (GraphPad Software), in which 1-way ANOVA with Bonferroni’s to the TDP-43 subtype) were made using single-variable (i.e., unad- post hoc test was performed for multiple group comparisons, and an justed) and multivariable linear regression models. Multivariable unpaired Student t test was performed when only 2 groups were com- models were adjusted for age at death, sex, and RIN, and truncated pared. The statistical test and number of independent experiments STMN2 RNA in the frontal cortex was examined on the base 2 log- used for each analysis are indicated in the figure legends. arithmic scale, owing to its skewed distribution. Regression coeffi- Study approval. The NYGC ALS Consortium samples presented in cients (referred to as β) and 95% CIs were estimated and interpreted this work were acquired through various IRB protocols from member as the difference in mean truncated STMN2 RNA levels in the frontal sites and the Target ALS postmortem tissue core and transferred to the cortex (on the base 2 logarithmic scale) between the given group of NYGC in accordance with all applicable foreign, domestic, federal, state, FTLD-TDP patients and the given reference group (cognitively nor- and local laws and regulations for processing, sequencing, and analyses. mal controls or patients with PSP). Additionally, to further assess the Postmortem brain tissues from patients with FTLD-TDP or PSP ability of truncated STMN2 RNA in the frontal cortex to discriminate and from cognitively normal individuals were obtained from the Mayo between FTLD-TDP and non–FTLD-TDP individuals (again, sep- Clinic Florida Brain Bank. Diagnosis was independently ascertained arately for cognitively normal controls and patients with PSP), we by trained neurologists and neuropathologists upon neurological and estimated the area under the ROC curve (AUC); an AUC of 0.5 corre- pathological examinations, respectively. Written informed consent sponded to the predictive ability equal to that of chance, whereas an was given by all participants or authorized family members, and all AUC of 1.0 indicated perfect predictive ability. Bonferroni’s correc- protocols were approved by the IRB and ethics committee of the Mayo tion was applied to adjust for the 5 different statistical tests that were Clinic. A summary of the patients’ characteristics is provided in Table 1.

6090 jci.org Volume 130 Number 11 November 2020 The Journal of Clinical Investigation RESEARCH ARTICLE

Sample size was determined by tissue availability and the quality of DSK, RCP, KAJ, and DWD provided human tissue samples as well RNA obtained from such samples, as described below. as pathological, genetic, and clinical information. All authors read Data and code availability. RNA-Seq data generated through the and approved the manuscript. NYGC ALS Consortium in this study can be accessed via the NCBI’s GEO database (GEO GSE137810, GSE124439, GSE116622, and Acknowledgments GSE153960). All RNA-Seq data generated by the NYGC ALS Consor- We thank the Target ALS Postmortem Core for providing post- tium are made immediately available to all members of the Consor- mortem brain samples, and we thank all the patients and their tium and with other consortia with whom we have a reciprocal sharing families for their contribution to this study. See Supplemental arrangement. To request immediate access to new and ongoing data Acknowledgments for consortium details. This work was sup- generated by the NYGC ALS Consortium and for samples provided ported by the NINDS, NIH (R35NS097273, P01NS084974, through the Target ALS Postmortem Core, complete a genetic data R21NS084528, and R01NS088689, to LP; P01NS099114, to TFG request form at [email protected]. and LP; R01AG037491, to KAJ; P50AG016574, to RCP and BFB; U01AG006786, to RCP; and U01AG045390, to BFB); the National Author contributions Institute on Aging (R56AG055824, to JH and TR); the Department MP, JH, and SP share co–first authorship. The order of their names of Defense (ALSRP AL130125, to LP); the Mayo Clinic Founda- was established on the basis of their individual contributions and tion (to LP); the Association of Frontotemporal Dementia (AFTD) the importance of those contributions to this study. MP led the (to LP); the Amyotrophic Lateral Sclerosis Association (to LP and human brain RNA and protein studies that provided the basis of MP); the Robert Packard Center for ALS Research at Johns Hop- the manuscript. JH led the RNA-Seq data analyses that comple- kins (to LP); Target ALS (to TFG and LP); the Canadian Institute mented the RNA/protein studies and increased the impact of this of Health Research (to SP); the UK Medical Research Council (to manuscript. SP led iPSC studies and was closely involved in the PF and EMCF); the Motor Neuron Disease Association (to PF); the RNA/protein studies as well. MP, JH, SP, TFG, PF, and LP designed Rosetrees Foundation (to PF and EMCF); the National Institute the study. MP, SP, ALB, CC, YS, MY, LMD, YC, KJW, CB, AC, KT, for Health Research (NIHR) University College London Hospitals BTS, KB, SC, and EMCF generated the iPSC lines and processed (UCLH) Biomedical Research Centre (to PF and ALB); Alzheimer’s iPSC and human tissue samples for qRT-PCR and RNA-Seq. SEH Research UK (to TL); and the UK Dementia Research Institute, and MW contributed to the experimental design for CRISPRi which receives its funding from DRI Ltd., supported by the UK knockdown of TARDBP in iPSC-derived neurons. In the NYGC Medical Research Council, the Alzheimer’s Society, and Alzhei- ALS Consortium, members contributed postmortem samples and mer’s Research UK (to BTS and SC). This work was also support- clinical information that were curated deidentified by DK. and ed in part by the Intramural Research Program, NINDS, NIH (to processed for RNA-Seq by IH, DC, and NP. NP and SF oversee MW). All NYGC ALS Consortium activities are supported by the Consortium resources and data distribution. DF and HP designed ALS Association (15-LGCA-234) and the Tow Foundation. the methodology, reviewed sample and data quality, and coordi- nated NYGC ALS Consortium postmortem RNA research activity. Address correspondence to: Leonard Petrucelli, Department of JH, YCW, PS, ALB, MS, and TR performed RNA-Seq data analy- Research, Neuroscience, Mayo Clinic College of Medicine, 4500 San ses. MP, JMK, JS, EAT performed NanoString assays and analyzed Pablo Road, Jacksonville, Florida 32224, USA. Phone: 904.953.2855; the data. SP and TFG performed p–TDP-43 immunoassays. MP, Email: [email protected]. Or to: Pietro Fratta, Depart- MGH, MRS, ALB, and MS performed statistical analyses. MP, JH, ment of Neuromuscular Diseases, UCL Queen Square Institute of SP, ALB, TFG, PF, and LP wrote the manuscript. CFDC, MD, SK, Neurology, Queen Square, London, WC1N 3BG, United Kingdom. JN, TL, the NYGC ALS Consortium, BO, MVB, RR, NRGR, BFB, Phone: 44.0.2034484112; Email: [email protected].

1. Ling SC, Polymenidou M, Cleveland DW. associated with TDP-43 depletion. BMC Med to frontotemporal dementia without motor neu- Converging mechanisms in ALS and FTD: dis- Genomics. 2017;10(1):38. ron disease. Hum Mutat. 2009;30(11):E974–83. rupted RNA and protein homeostasis. Neuron. 7. Klim JR, et al. ALS-implicated protein TDP- 13. Mackenzie IR, et al. Dipeptide repeat protein 2013;79(3):416–438. 43 sustains levels of STMN2, a mediator of pathology in C9ORF72 mutation cases: clini- 2. Polymenidou M, et al. Long pre-mRNA depletion motor neuron growth and repair. Nat Neurosci. co-pathological correlations. Acta Neuropathol. and RNA missplicing contribute to neuronal 2019;22(2):167–179. 2013;126(6):859–879. vulnerability from loss of TDP-43. Nat Neurosci. 8. Melamed Z, et al. Premature polyadenyla- 14. Neumann M, et al. Ubiquitinated TDP-43 in fron- 2011;14(4):459–468. tion-mediated loss of stathmin-2 is a hallmark totemporal lobar degeneration and amyotrophic 3. Tollervey JR, et al. Characterizing the RNA tar- of TDP-43-dependent neurodegeneration. Nat lateral sclerosis. Science. 2006;314(5796):130–133. gets and position-dependent splicing regulation Neurosci. 2019;22(2):180–190. 15. Sampathu DM, et al. Pathological heterogeneity by TDP-43. Nat Neurosci. 2011;14(4):452–458. 9. Shin JE, Geisler S, DiAntonio A. Dynamic regula- of frontotemporal lobar degeneration with ubiq- 4. Ling JP, Pletnikova O, Troncoso JC, Wong PC. tion of SCG10 in regenerating axons after injury. uitin-positive inclusions delineated by ubiquitin TDP-43 repression of nonconserved cryptic Exp Neurol. 2014;252:1–11. immunohistochemistry and novel monoclonal exons is compromised in ALS-FTD. Science. 10. Neumann M, Mackenzie IRA. Review: Neuropa- antibodies. Am J Pathol. 2006;169(4):1343–1352. 2015;349(6248):650–655. thology of non-tau frontotemporal lobar degenera- 16. Davidson Y, et al. Ubiquitinated pathological 5. Jeong YH, et al. Tdp-43 cryptic exons are highly tion. Neuropathol Appl Neurobiol. 2019;45(1):19–40. lesions in frontotemporal lobar degeneration variable between cell types. Mol Neurodegener. 11. Sreedharan J, et al. TDP-43 mutations in familial contain the TAR DNA-binding protein, TDP-43. 2017;12(1):13. and sporadic amyotrophic lateral sclerosis. Sci- Acta Neuropathol. 2007;113(5):521–533. 6. Humphrey J, Emmett W, Fratta P, Isaacs AM, Pla- ence. 2008;319(5870):1668–1672. 17. Arai T, et al. TDP-43 is a component of ubiqui- gnol V. Quantitative analysis of cryptic splicing 12. Borroni B, et al. Mutation within TARDBP leads tin-positive tau-negative inclusions in fronto-

jci.org Volume 130 Number 11 November 2020 6091 RESEARCH ARTICLE The Journal of Clinical Investigation

temporal lobar degeneration and amyotrophic pathol. 2017;134(2):241–254. reproducible computational workflows. Nat Bio- lateral sclerosis. Biochem Biophys Res Commun. 35. Bang J, Spina S, Miller BL. Frontotemporal technol. 2017;35(4):316–319. 2006;351(3):602–611. dementia. Lancet. 2015;386(10004):1672–1682. 54. Bolger AM, Lohse M, Usadel B. Trimmomatic: 18. Hatanpaa KJ, et al. TAR DNA-binding protein 43 36. Wang Q, et al. The landscape of multiscale tran- a flexible trimmer for Illumina sequence data. immunohistochemistry reveals extensive neu- scriptomic networks and key regulators in Par- Bioinformatics. 2014;30(15):2114–2120. ritic pathology in FTLD-U: a midwest-southwest kinson’s disease. Nat Commun. 2019;10(1):5234. 55. Frankish A, et al. GENCODE reference annota- consortium for FTLD study. J Neuropathol Exp 37. Amador-Ortiz C, et al. TDP-43 immunoreactivity tion for the human and mouse genomes. Nucleic Neurol. 2008;67(4):271–279. in hippocampal sclerosis and Alzheimer’s dis- Acids Res. 2019;47(D1):D766–D73. 19. Josephs KA, Stroh A, Dugger B, Dickson DW. ease. Ann Neurol. 2007;61(5):435–445. 56. Li B, Dewey CN. RSEM: accurate transcript quanti- Evaluation of subcortical pathology and clinical 38. Arai T, et al. Phosphorylated TDP-43 in Alzhei- fication from RNA-Seq data with or without a refer- correlations in FTLD-U subtypes. Acta Neuro- mer’s disease and dementia with Lewy bodies. ence genome. BMC Bioinformatics. 2011;12:323. pathol. 2009;118(3):349–358. Acta Neuropathol. 2009;117(2):125–136. 57. Li H, et al. The Sequence Alignment/ 20. Mackenzie IR, et al. A harmonized classification 39. Higashi S, et al. Concurrence of TDP-43, tau and Map format and SAMtools. Bioinformatics. system for FTLD-TDP pathology. Acta Neuro- alpha-synuclein pathology in brains of Alzhei- 2009;25(16):2078–2079. pathol. 2011;122(1):111–113. mer’s disease and dementia with Lewy bodies. 58. [No authors listed]. Picard Toolkit. Broad Insti- 21. Lee EB, et al. Expansion of the classification of Brain Res. 2007;1184:284–294. tute. http://broadinstitute.github.io/picard/. FTLD-TDP: distinct pathology associated with 40. Hu WT, et al. Temporal lobar predominance of TDP- Accessed September 3, 2020. rapidly progressive frontotemporal degeneration. 43 neuronal cytoplasmic inclusions in Alzheimer 59. Ewels P, Magnusson M, Lundin S, Kaller M. Mul- Acta Neuropathol. 2017;134(1):65–78. disease. Acta Neuropathol. 2008;116(2):215–220. tiQC: summarize analysis results for multiple tools 22. Mackenzie IR, et al. Heterogeneity of ubiquitin 41. Josephs KA, et al. Abnormal TDP-43 immu- and samples in a single report. Bioinformatics. pathology in frontotemporal lobar degeneration: noreactivity in AD modifies clinicopatho- 2016;32(19):3047–3048. classification and relation to clinical phenotype. logic and radiologic phenotype. Neurology. 60. R Core Team. R: A language and environment for Acta Neuropathol. 2006;112(5):539–549. 2008;70(19 Pt 2):1850–1857. statistical computing. R Foundation for Statistical 23. Tan RH, et al. Classification of FTLD-TDP cases 42. Kadokura A, et al. Phosphorylation-dependent Computing, Vienna, Austria. http://www.R-proj- into pathological subtypes using antibodies TDP-43 antibody detects intraneuronal dot-like ect.org/. Accessed September 3, 2020. against phosphorylated and non-phosphorylated structures showing morphological characters 61. Wickham H. tidyverse: Easily Install and Load the TDP43. Acta Neuropathol Commun. 2013;1:33. of granulovacuolar degeneration. Neurosci Lett. ‘Tidyverse’. https://cran.r-project.org/web/pack- 24. Josephs KA, et al. Neuropathological background 2009;463(1):87–92. ages/tidyverse/index.html. Published November of phenotypical variability in frontotemporal 43. Kadokura A, Yamazaki T, Lemere CA, Takatama 21, 2019. Accessed September 3, 2020. dementia. Acta Neuropathol. 2011;122(2):137–153. M, Okamoto K. Regional distribution of TDP-43 62. Kassambara A. ggpubr: ‘ggplot2’ Based Publica- 25. Josephs KA, et al. Neuropathologic features of inclusions in Alzheimer disease (AD) brains: tion Ready Plots. Github. https://github.com/kas- frontotemporal lobar degeneration with ubiq- their relation to AD common pathology. Neuropa- sambara/ggpubr. Accessed Septmeber 3, 2020. uitin-positive inclusions with progranulin gene thology. 2009;29(5):566–573. 63. Pedersen TL. patchwork: the composer of plots. (PGRN) mutations. J Neuropathol Exp Neurol. 44. Uryu K, et al. Concomitant TAR-DNA-binding R Project. https://cran.r-project.org/web/pack- 2007;66(2):142–151. protein 43 pathology is present in Alzheimer ages/patchwork/index.html. Published June 22, 26. Mackenzie IR, et al. The neuropathology of disease and corticobasal degeneration but not 2020. Accessed September 3, 2020. frontotemporal lobar degeneration caused in other tauopathies. J Neuropathol Exp Neurol. 64. Aguet F, et al. The GTEx Consortium atlas of genet- by mutations in the progranulin gene. Brain. 2008;67(6):555–564. ic regulatory effects across human tissues. bioRxiv. 2006;129(Pt 11):3081–3090. 45. Josephs KA, et al. Rates of hippocampal atrophy https://doi.org/10.1101/787903. Published Octo- 27. Feneberg E, Gray E, Ansorge O, Talbot K, Turner and presence of post-mortem TDP-43 in patients ber 3, 2019. Accessed September 3, 2020. MR. Towards a TDP-43-based biomarker for ALS with Alzheimer’s disease: a longitudinal retrospec- 65. Schroeder A, et al. The RIN: an RNA integrity and FTLD. Mol Neurobiol. 2018;55(10):7789–7801. tive study. Lancet Neurol. 2017;16(11):917–924. number for assigning integrity values to RNA 28. Nana AL, et al. Neurons selectively target- 46. Josephs KA, et al. TAR DNA-binding protein measurements. BMC Mol Biol. 2006;7:3. ed in frontotemporal dementia reveal early 43 and pathological subtype of Alzheimer’s 66. Jaffe AE, et al. qSVA framework for RNA quality stage TDP-43 pathobiology. Acta Neuropathol. disease impact clinical features. Ann Neurol. correction in differential expression analysis. 2019;137(1):27–46. 2015;78(5):697–709. Proc Natl Acad Sci U S A. 2017;114(27):7130–7135. 29. Koga S, et al. Corticobasal degeneration with 47. Chen S, Zhou Y, Chen Y, Gu J. fastp: an ultra-fast 67. Koster J, Rahmann S. Snakemake-a scalable TDP-43 pathology presenting with progres- all-in-one FASTQ preprocessor. Bioinformatics. bioinformatics workflow engine. Bioinformatics. sive supranuclear palsy syndrome: a distinct 2018;34(17):i884–i90. 2012;28(19):2520–2522. clinicopathologic subtype. Acta Neuropathol. 48. Dobin A, et al. STAR: ultrafast universal RNA-seq 68. Humphrey J. STMN2 splicing pipeline. https:// 2018;136(3):389–404. aligner. Bioinformatics. 2013;29(1):15–21. github.com/RajLabMSSM/STMN2-splicing. 30. Tian R, et al. CRISPR interference-based platform 49. Liao Y, Smyth GK, Shi W. featureCounts: an Accessed September 3, 2020. for multimodal genetic screens in human iPSC-­ efficient general purpose program for assigning 69. Feng YY, , et al. RegTools: Integrated analysis of derived neurons. Neuron. 2019;104(2):239–255.e12. sequence reads to genomic features. Bioinformatics. genomic and transcriptomic data for discovery of 31. Fratta P, et al. Mice with endogenous TDP-43 2014;30(7):923–930. splicing variants in cancer. bioRxiv. https://doi. mutations exhibit gain of splicing function and 50. Robinson MD, McCarthy DJ, Smyth GK. edgeR: org/10.1101/436634. Published November 25, characteristics of amyotrophic lateral sclerosis. a Bioconductor package for differential expres- 2018. Accessed September 3, 2020. EMBO J. 2018;37(11):e98684. sion analysis of digital gene expression data. 70. Li YI, et al. Annotation-free quantification of 32. Jiang J, et al. Gain of toxicity from ALS/ Bioinformatics. 2010;26(1):139–140. RNA splicing using LeafCutter. Nat Genet. FTD-linked repeat expansions in C9ORF72 51. Katz Y, Wang ET, Airoldi EM, Burge CB. Analysis 2018;50(1):151–158. is alleviated by antisense oligonucleotides tar- and design of RNA sequencing experiments for 71. Jackson JL, et al. Elevated methylation levels, geting GGGGCC-containing RNAs. Neuron. identifying isoform regulation. Nat Methods. reduced expression levels, and frequent contrac- 2016;90(3):535–550. 2010;7(12):1009–1015. tions in a clinical cohort of C9orf72 expansion 33. Prudencio M, et al. Repetitive element transcripts 52. Tam OH, et al. Postmortem cortex samples carriers. Mol Neurodegener. 2020;15(1):7. are elevated in the brain of C9orf72 ALS/FTLD identify distinct molecular subtypes of ALS: 72. Dafinca R, et al. Impairment of mitochondrial patients. Hum Mol Genet. 2017;26(17):3421–3431. retrotransposon activation, oxidative stress, and calcium buffering links mMutations in C9ORF72 34. Schludi MH, et al. Spinal poly-GA inclusions in a activated glia. Cell Rep. 2019;29(5):1164–1177. and TARDBP in iPS-derived motor neurons C9orf72 mouse model trigger motor deficits and 53. Di Tommaso P, Chatzou M, Floden EW, Barja from patients with ALS/FTD. Stem Cell Reports. inflammation without neuron loss. Acta Neuro- PP, Palumbo E, Notredame C. Nextflow enables 2020;14(5):892–908.

6092 jci.org Volume 130 Number 11 November 2020